Function gdn_decode_f32_norm_snap 

pub fn gdn_decode_f32_norm_snap(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    h_state: DevicePtr,
    query: DevicePtr,
    key: DevicePtr,
    value: DevicePtr,
    gate: DevicePtr,
    beta: DevicePtr,
    z_gate: DevicePtr,
    norm_weight: DevicePtr,
    output: DevicePtr,
    h_inter: DevicePtr,
    batch_size: u32,
    num_k_heads: u32,
    num_v_heads: u32,
    k_dim: u32,
    v_dim: u32,
    eps: f32,
    stream: u64,
) -> Result<()>

2026-09-25: super::gdn_decode_f32_norm plus an inline h-state snapshot.

h_inter receives the updated H after the state-norm clamp, the same values left in h_state; NULL skips the snapshot. Same grid, block and argument order as the parent, with h_inter appended.